A Service of
Leibniz-Informationszentrum econstor Wirtschaft Leibniz Information Centre Make Your Publications Visible. zbw for Economics
Blau, Francine D.; Kahn, Lawrence M.
Working Paper The Gender Wage Gap: Extent, Trends, and Explanations
IZA Discussion Papers, No. 9656
Provided in Cooperation with: IZA – Institute of Labor Economics
Suggested Citation: Blau, Francine D.; Kahn, Lawrence M. (2016) : The Gender Wage Gap: Extent, Trends, and Explanations, IZA Discussion Papers, No. 9656, Institute for the Study of Labor (IZA), Bonn
This Version is available at: http://hdl.handle.net/10419/130341
Standard-Nutzungsbedingungen: Terms of use:
Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Documents in EconStor may be saved and copied for your Zwecken und zum Privatgebrauch gespeichert und kopiert werden. personal and scholarly purposes.
Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle You are not to copy documents for public or commercial Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich purposes, to exhibit the documents publicly, to make them machen, vertreiben oder anderweitig nutzen. publicly available on the internet, or to distribute or otherwise use the documents in public. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, If the documents have been made available under an Open gelten abweichend von diesen Nutzungsbedingungen die in der dort Content Licence (especially Creative Commons Licences), you genannten Lizenz gewährten Nutzungsrechte. may exercise further usage rights as specified in the indicated licence. www.econstor.eu IZA DP No. 9656
The Gender Wage Gap: Extent, Trends, and Explanations
Francine D. Blau Lawrence M. Kahn
January 2016 DISCUSSION PAPER SERIES
Forschungsinstitut zur Zukunft der Arbeit Institute for the Study of Labor
The Gender Wage Gap: Extent, Trends, and Explanations
Francine D. Blau Cornell University, NBER, IZA, DIW and CESifo
Lawrence M. Kahn Cornell University, IZA, CESifo and NCER
Discussion Paper No. 9656 January 2016
IZA
P.O. Box 7240 53072 Bonn Germany
Phone: +49-228-3894-0 Fax: +49-228-3894-180 E-mail: [email protected]
Any opinions expressed here are those of the author(s) and not those of IZA. Research published in this series may include views on policy, but the institute itself takes no institutional policy positions. The IZA research network is committed to the IZA Guiding Principles of Research Integrity.
The Institute for the Study of Labor (IZA) in Bonn is a local and virtual international research center and a place of communication between science, politics and business. IZA is an independent nonprofit organization supported by Deutsche Post Foundation. The center is associated with the University of Bonn and offers a stimulating research environment through its international network, workshops and conferences, data service, project support, research visits and doctoral program. IZA engages in (i) original and internationally competitive research in all fields of labor economics, (ii) development of policy concepts, and (iii) dissemination of research results and concepts to the interested public.
IZA Discussion Papers often represent preliminary work and are circulated to encourage discussion. Citation of such a paper should account for its provisional character. A revised version may be available directly from the author. IZA Discussion Paper No. 9656 January 2016
ABSTRACT
The Gender Wage Gap: Extent, Trends, and Explanations*
Using PSID microdata over the 1980-2010, we provide new empirical evidence on the extent of and trends in the gender wage gap, which declined considerably over this period. By 2010, conventional human capital variables taken together explained little of the gender wage gap, while gender differences in occupation and industry continued to be important. Moreover, the gender pay gap declined much more slowly at the top of the wage distribution that at the middle or the bottom and by 2010 was noticeably higher at the top. We then survey the literature to identify what has been learned about the explanations for the gap. We conclude that many of the traditional explanations continue to have salience. Although human capital factors are now relatively unimportant in the aggregate, women’s work force interruptions and shorter hours remain significant in high skilled occupations, possibly due to compensating differentials. Gender differences in occupations and industries, as well as differences in gender roles and the gender division of labor remain important, and research based on experimental evidence strongly suggests that discrimination cannot be discounted. Psychological attributes or noncognitive skills comprise one of the newer explanations for gender differences in outcomes. Our effort to assess the quantitative evidence on the importance of these factors suggests that they account for a small to moderate portion of the gender pay gap, considerably smaller than say occupation and industry effects, though they appear to modestly contribute to these differences.
JEL Classification: J16, J24, J31, J71
Keywords: gender, gender pay gap, wage differentials, discrimination, human capital investment, occupations, occupational segregation
Corresponding author:
Francine D. Blau ILR School Cornell University 268 Ives Hall Ithaca, NY 14853-3901 USA E-mail: [email protected]
* Forthcoming in the Journal of Economic Literature. We are indebted to Steven Durlauf, Janet Currie, Anne Winkler, Martha Bailey, and four anonymous referees for helpful comments and suggestions and to Jason Cook and Ankita Patnaik for excellent research assistance. 1. Introduction The gender wage gap has now been intensively investigated for a number of decades, but also remains an area of active and innovative research. In this article, we provide new empirical estimates delineating the extent of and trends in the gender wage gap and their potential explanations. We then survey the literature to identify what has been learned about the explanations of the gap, both those that can be readily included in conventional analyses and those that cannot; both traditional explanations and newer ones that have been offered. Our primary focus will be on the United States, although we also place the United States in a comparative perspective, particularly as such comparisons help to further our understanding of the sources of the gender wage gap. The focus on the United States is in part designed to make our task more manageable, as there has been an explosion of research on this topic across many countries. Nonetheless, we believe much of what we have learned for the United States is applicable to other countries, particularly other economically advanced nations. In our comprehensive review of the literature, we particularly emphasize areas where there has been exciting new research on more traditional explanations and on newer explanations and trends, including research on gender differences in psychological attributes/noncognitive skills and mathematics test scores, and on the reversal of the gender education gap. The long-term trend has been a substantial reduction in the gender wage gap, both in the United States and in other economically advanced nations (Blau and Kahn 2008). However, the shorter term picture in the United States has been somewhat mixed. The period of strongest wage convergence between men and women was the 1980s, and progress has been slower and more uneven since then. Moreover, a number of other related trends appear to have plateaued or slowed since the 1990s, including increases in female labor force participation rates and reductions in occupational segregation by sex. The plan of the paper is as follows. In Section 2, we begin by documenting the changes in the gender gap that have occurred in the United States since the 1950s based on published data. We then provide new analyses for the 1980 to 2010 period that include decompositions of the changes in the gender wage gap into portions associated with key characteristics such as schooling, experience, industry, occupation and union status. We also examine how women fared relative to men at various points in the wage distribution. Our decompositions show the importance of these measured factors in accounting for the levels and changes in the gender pay gap. We also find that an unexplained gap remains and, moreover, that it has been stable subsequent to a dramatic narrowing over the 1980s. In the remaining sections we probe what is known about the various factors that contribute to the gender pay gap, including the extent of and trends in these factors. Some of the variables we consider are measured in our data set and included in our analysis in Section 2, as well as other similar type analyses. Other factors are not included and presumably help to provide insight into the sources of the unexplained gap. However, it is important to point out that the effects of factors that are not explicitly included in traditional regression analyses may be taken into account to some extent by measured variables. For example, women have been found to be more risk averse than men on average which could lower their relative wages. However, to the extent that this factor operates through gender differences in occupational sorting, e.g., if it results in women avoiding occupations with greater variance in earnings, regression analyses that control for occupation will adjust for this factor. Our consideration of explanatory factors begins in Section 3 where we discuss variables economists have traditionally emphasized in studying the gender pay gap. These include human capital (schooling and work experience), the family division of labor, compensating wage differentials, discrimination, and issues relating to selection into the labor force.1 Gender differences in occupations,
1 Selection issues arise because we do not observe wage offers for people who are not currently employed and a smaller share of the female than of the male population is employed. Moreover, the share of both groups, but particularly of women, who are employed has changed over time.
1 industries and firms are a component of this discussion. We especially emphasize new empirical and theoretical research on these traditional factors. We then turn in Section 4 to a discussion of a relatively new field of research among economists studying gender: the impact of norms, psychological attributes and noncognitive skills on the gender pay gap. This body of work includes both survey evidence and lab and field experiments. It has the potential to help explain not only what economists have called the unexplained gender wage gap (i.e. the portion not accounted for gender differences in measured qualifications) but also gender differences in some of the measured factors themselves. However, a theme that emerges from some of the experimental work is that some psychological attitudes may themselves be influenced by context. For example, anticipated treatment of women in the labor market may affect their aspirations. The formation of norms and attitudes thus in our view is a potentially fruitful area of research that has received relatively little attention by economists. We then turn in Section 5 to a discussion of the impact of policy on the gender wage gap, including both antidiscrimination policy and family leave policies. While the discussion up to this point emphasizes gender-specific factors (i.e., gender differences in behavior, qualifications, and treatment), in Section 6, we highlights that the overall structure of wages can affect the gender wage gap, given that men and women have different skills and qualifications and work in different occupations and industries. Hence, changes over time or differences across countries in the return to various skills or to working in high-paying sectors (occupations or industries) will affect the gender pay gap. As another example, policies such as minimum wages or union negotiated wage floors that bring up the bottom of the distribution will disproportionately affect women even if the law or union agreement is not gender- specific. In Section 6, we discuss wage structure and refer to evidence both in the United States and from other countries in which the wage structure is much more compressed as a result of union wage-setting. Finally, Section 7 presents conclusions. 2. Overview of the US Gender Wage Gap In this section, we use published data, information from the Michigan Panel Study of Income Dynamics (PSID), and the March Current Population Survey (CPS) to establish the facts on the levels and trends in the US gender wage gap and on their sources (in a descriptive sense). Accounting for the sources of the level and changes in the gender pay gap will provide guidance for understanding recent research studying gender and the labor market. Figure 1 shows the long-run trends in the gender pay gap over the 1955-2014 period based on two published series: usual weekly earnings of full-time workers and annual earnings of full-time, year-round workers. After many years with a stable female/male earnings ratio of roughly 60%, women’s relative wages began to rise sharply in the 1980s, with a continued, but slower and more uneven rate of increase thereafter. By 2014, women full-time workers earned about 79% of what men did on an annual basis and about 83% on a weekly basis. To better understand the sources of the gender wage gap, we analyze data from the PSID, which is the only data source that has information on actual labor market experience (a crucial variable in gender analyses) for the full age range of the population. We focus on men and women age 25-64 who were full- time, non-farm, wage and salary workers and who worked at least 26 weeks during the preceding year. The focus on full-time workers and those with substantial labor force attachment over the year is designed to identify female and male workers with fairly similar levels of labor market commitment. However, we have repeated our analyses on the full sample of all wage and salary earners (including those employed part time or part year) and obtained very similar results to those shown here. The sample is also restricted to family heads and spouses/cohabitors because the PSID only supplies the crucial work history information for these individuals. Due to this and other limitations in coverage by the PSID, described in the Data Appendix, we present some additional data on the gender pay gap using the fully nationally
2 representative March CPS.2 The empirical results in this section are of interest in and of themselves and also serve to set the stage for the literature review to follow by providing a frame of reference for how each of the measured factors discussed relates to the overall gender wage gap and changes in the gap. Our data cover the 1980-2010 period, in which, as Figure 1 shows, women have made major gains in relative wages. Table 1 shows the evolution of the female-male ratio of average hourly earnings at the mean and also at the 10th, 50th, and 90th percentiles for four years—1980, 1989, 1998, and 2010—based on both PSID and CPS data.3 Because i earnings refer to the previous year, we use, for example, the 1981 data to measure wages in 1980. The overall pattern is very similar across the two data sets, and also largely matches that in the published data shown in Figure 1, increasing one’s confidence in the PSID.4 Specifically, gains in the female/male wage ratio were largest in the 1980s and occurred at a slower pace thereafter, with the ratio rising from 62-64% in 1980 to 72-74% in 1989, with a further increase to 79- 82% by 2010.5 The time pattern at the bottom (10th percentile), middle (50th percentile) and top (90th percentile) of the wage distribution is similar to that for the overall mean: the gender wage ratio rose over the period, with the largest gains during the 1980s. However, a closer examination shows that women gained least, in a relative sense, at the top. In both the PSID and CPS, women at the top had a slightly higher pay ratio than those in the middle and a slightly lower pay ratio than those at the bottom in 1980. Yet by 2010, in both data sets, women’s relative pay at the top was considerably less than that at the middle and bottom of the distribution: 8-9 percentage points less than that at the middle or bottom in the PSID, and 6-11 percentage points less in the CPS. Later in this section, we will consider the role of measured factors in accounting for the slower reduction at the top and in following sections we will attempt to shed additional light by reviewing the literature on the labor market for highly skilled workers. At the same time that the gender pay gap has been narrowing, women have been increasing their relative labor market qualifications and commitment to work. Tables 2 and 3 show the extent of such changes among our PSID sample of full-time workers. Table 2 focuses on the prime human capital determinants of men’s and women’s wages, education and actual full-time experience. In the case of education, there was a dramatic reversal of the gender gap. In 1981, women had lower average levels of schooling than men and were less likely to have exactly a bachelor’s or an advanced degree. Over the period, women narrowed the education gap with men and, by 2011, women had higher average levels of schooling and were more likely to have an advanced degree than men.6 While men had a slightly higher
2 Additional information on the details of our data preparation and analysis is available in the online Data Appendix. Means and other data presented here are for the sample used in our regression analyses. In the PSID, we exclude cases with missing data on the dependent or explanatory variables, or variables needed to construct them. In the CPS, we exclude cases with allocated earnings. See Table 1 for sample sizes. 3 Entries are calculated as exp (D), where D is the female log wage at the mean, or at the indicated percentile, minus the corresponding male log wage. 4 The unemployment rate was 7.1% in 1980, 5.3% in 1989, 4.5% in 1998, and 9.6% in 2010 (see http://data.bls.gov/timeseries/LNU04000000?years_option=all_years&periods_option=specific_periods&periods=A nnual+Data , accessed December 27, 2015). The high level of unemployment in 2010 may raise concerns about the representativeness of that year for studying the gender pay gap. Reassuringly, however, we found similar results when we ended our PSID sample in 2006, before the Great Recession began. 5 The larger female gains in relative wages during 1980s is a result we have studied in some detail in prior work (Blau and Kahn 2006), where we explicitly compared the 1980s and the 1990s. 6 Tables 2 and 3 refer to 1981, 1990, 1999, and 2011 rather than 1980, etc., as shown in Table 1, because earnings refer to the previous year, while other variables are measured as of the survey date.
3 incidence of having exactly a bachelor’s degree, women were more likely to have at least a bachelor’s degree (i.e. the sum of the Bachelor’s Degree Only and Advanced Degree categories).7 In the case of labor market experience, the story is one of a substantial narrowing of the gender experience gap. In 1981, men had nearly 7 more years of full-time labor market experience on average than women. By 2011, the gap had fallen markedly to only 1.4 years, with the fastest rate of increase in women’s relative experience occurring during the 1980s.8 Thus, on these two basic measures of human capital—schooling and actual labor market experience—women made important gains during the 1981- 2011 period, reversing the education gap and greatly reducing the experience gap. Table 3 further explores trends in the determinants of wages by showing gender differences in the incidence of high-level jobs as well as collective bargaining coverage. Rising employment in managerial or professional jobs may be an indicator of increasing human capital or work commitment, even controlling for levels of schooling and actual labor market experience. For example, such jobs may entail higher levels of responsibility and pressure than other jobs, and only those with the appropriate training and commitment may be qualified to take them. Increases in women’s relative representation in such jobs may then be a further indicator of their rising human capital and labor market commitment. However, women’s representation in such jobs may also be affected by employer discrimination in entry or promotions. Women’s improvements may therefore also reflect reductions in discrimination. Both interpretations are plausible. First, it seems likely that women’s increasing levels of schooling and, as discussed below, increasing representation in lucrative fields of study, as well as their rising experience levels would be expected to lead to their greater representation in high-level positions. Second, given women’s increasing qualifications and commitment to the labor market, employer incentives for statistical discrimination (this concept is discussed further below) have likely been reduced. Under either interpretation, studying these differences can yield insights into the sources of the gender pay gap. Table 3 shows remarkable increases in women’s relative representation in such high- level jobs. The male advantage in managerial jobs fell from 12 percentage points in 1981 to just two percentage points in 2011. Moreover, while women were more likely than men to work in professional jobs throughout the period, their advantage grew from five percentage points in 1981 to nine percentage points in 2011. However, many women in professional jobs remain employed in traditionally female occupations such as nursing or K-12 teaching that are generally less lucrative than traditionally male professions. We therefore also show in Table 3 gender differences in the incidence of employment in “male” professional jobs, which we define as professional jobs other than nursing or K-12 and other non- college teaching positions, most of which were predominantly male at the start of our period. While men were four percentage points more likely than women to be in such jobs in 1981, by 2011, the gender gap had been virtually eliminated. At the same time women were making these occupational gains, they were greatly reducing their concentration in administrative support and clerical jobs.9 In addition to these occupational changes, one notable feature of the post-1980 labor market is the steady reduction in the portion of the economy covered by collective bargaining. Table 3 shows that this
7 CPS data also show that, in 1981, men had higher levels of schooling and incidence of bachelor’s or advanced degrees than women; by 2011, women in the CPS had higher levels of schooling than men, as in the PSID. However, in the 2011 CPS, women not only had a higher incidence of advanced degrees, but also a slightly higher incidence of exactly a college degree than men. 8 Some of the small experience gap in 2011 may have resulted from the recession. For example, in 2007 (i.e. before the recession), the full time experience gap was 2.6 years, compared to 2011’s gap of 1.4 years and the 1999 gap of 3.8 years. Whether the fall to 1.4 years by 2011 was a continuation of a trend or was due to the recession is unclear, though the upshot is the same: a substantial reduction in the gender experience gap. 9 We obtained very similar results on the gender gaps in managerial, professional, and “male” professional employment using the March CPS.
4 reduction hit men much harder than women. Specifically, men’s collective bargaining coverage fell from 34% in 1981 to 17% in 2011, while women’s coverage only declined from 21% to 19%.10 As is the case with women’s gains in education, full time labor market experience, and employment in high-level occupations, we expect the elimination of the gender gap in collective bargaining coverage to contribute to a reduction in the gender pay gap.11 How have gender differences in women’s labor market qualifications and employment location affected the gender wage gap? And how have improvements in women’s relative characteristics affected changes in the gender wage gap? We study these questions by decomposing levels and changes in the gender wage gap over the 1980-2010 period using log wage regressions. We proceed in two stages. First, we estimate wage models that only control for education, experience, race/ethnicity, region, and metropolitan area residence. We term this the “human capital specification,” since other than basic controls, we include only human capital variables—education and experience. Second, we augment this model with a series of industry, occupation and union coverage dummy variables. We term this equation the “full specification.” Because these latter variables may have an ambiguous interpretation—i.e., they may represent human capital, other labor market skills, and commitment, on the one hand, or employer discrimination, on the other hand—we present both versions. Note that we do not control for marital status or number of children, since these are likely to be endogenous with respect to women’s labor force decisions. Our decompositions can be viewed as reduced forms with respect to family formation decisions.12 We measure education by controlling for years of schooling, plus dummy variables for having exactly a bachelor’s degree and an advanced degree. We include measures of both full-time and part-time labor market experience and their squares. Race and ethnicity are controlled for using four mutually- exclusive categories: white non-Hispanic (the excluded category), black non-Hispanic, other non- Hispanic, and Hispanic. We control for three of the four Census regions as well as including a dummy variable for residence in a metropolitan area. In the full specification, we additionally control for a series of 14 industry and 20 occupation dummy variables, government employment, and a collective bargaining coverage dummy variable. (In the decompositions below, government employment is included with industry.) The construction of these categories took account of changes in the PSID’s coding scheme over the period and is described in the online Data Appendix. 2.1 Explaining the Gender Wage Gap at the Mean Figure 2 shows female to male log wage ratios, (i) unadjusted for covariates (i.e. reproduced from Table 1), (ii) adjusted for the covariates in the human capital specification, and (iii) adjusted for the covariates in the full specification. The adjusted female/male wage ratios shown in Figure 2 and analyzed in more detail in Table 4 are computed using a traditional Oaxaca-Blinder decomposition of male-female differences in log wages into a component accounted for by differences in characteristics and an
10 While the PSID data show women as now having slightly higher collective bargaining coverage than men, US Bureau of Labor Statistics data show men continuing to retain a small edge. Specifically, in 1983, among those 16 years and older, 27.7% of men were covered by collective bargaining, compared to 18.0% of women; by 2011, men’s coverage had decreased to 13.5%, while women’s declined to 12.5% (http://data.bls.gov/pdq/SurveyOutputServlet , accessed August 18, 2014). 11 In the PSID, the convergence in the collective bargaining coverage of men and women was a result of both a larger fall in men’s private sector coverage and an increase in women’s public sector coverage, with men’s public sector coverage remaining stable. 12 An additional reason we did not control for marital status and children in our basic regressions is that such variables are expected to increase male wages but to decrease female wages, complicating one’s assessment of gender gaps in explanatory variables. Nonetheless, when we included these variables in our basic wage regressions, the decomposition results were very similar to those shown here.
5 unexplained component (Oaxaca 1973, Blinder 1973). The latter is often taken to be an estimate of the extent of discrimination—i.e., unequal pay for equally qualified workers. However, the unexplained portion of the gender pay gap may include the effects of unmeasured productivity or compensating differentials, and some of the explanatory variables such as industry or occupation may be affected by discrimination. We consider this issue in greater detail in Section 3.9, while our discussion of research on selection, unmeasured attributes such as competitiveness or risk aversion, and possible glass ceilings will shed light on some possible sources of the pay gap that cannot be explained by measured characteristics. The following equations illustrate the Blinder Oaxaca decomposition. For year t, estimate separate male (m) and female (f) Ordinary Least Squares (OLS) wage regressions for individual i (the i and t subscripts are suppressed to simplify the notation):
(1)
(2) where is the log of wages, is a vector of explanatory variables such as education and experience, is a vector of coefficients and u is an error term.
Let and be respectively the OLS estimates of and , and denote mean values with a bar over the variable. Then, since OLS with a constant term produces residuals with a zero mean, we have:
(3)
The first term on the far right hand side of (3) is the impact of gender differences in the explanatory variables evaluated using the male coefficients. The second term is the unexplained differential and corresponds to the average female residual from the male wage equation. In Figure 2, we take the exponential of this residual and obtain the simulated female to male wage ratio, controlling for the indicated variables. This residual corresponds to an experiment where we take one woman, given her characteristics, and reward her according to the male reward system. One might think of such an experiment as the outcome of a discrimination case in which a firm that previously was found to have discriminated against women is now required to treat women the same as it treats men. The decomposition in (3) of course could be performed using the female coefficients and the male means, and we have performed such a decomposition as well, with similar results to the ones reported here, although the unexplained residual was somewhat larger using the male means.13 The results for the unadjusted ratios in Figure 2 mirror the trends from the published data, showing a large increase in the female-to-male wage ratio over the 1980s, with continued but smaller gains in subsequent decades.14 Over the 1980-2010 period as a whole, the unadjusted ratio increased
13 Some have argued that a wage regression pooling men and women should be used since it is claimed that this would be the wage regression prevailing in a nondiscriminatory labor market (Cotton 1988; Neumark 1988). We have not done so here because there would likely be general equilibrium changes if discrimination were eradicated, and we do not know what the resulting reward structure would look like. Instead, we take the more modest approach of performing the decomposition using alternative weights and comparing the results. As just mentioned, however, the experiment of taking a women and valuing her characteristics using the male coefficients does correspond to a real-life scenario. We should also point out that in data sets such as the CPS that do not measure actual experience, the female equation will give a less accurate estimate of the return to labor market experience than the male equation. 14 The US labor force aged over the 1980-2010 period, and it is well known that gender pay gaps increase with age. To investigate whether aging has influenced our picture of the trends in the gender wage gap, we re-weighted our data with 1980 age weights using a quartic in age in a procedure based on DiNardo, Fortin and Lemieux (1996). Men and women in our wage samples were in fact 3-4 years older in 2010 than 1980. When we repeated our
6 substantially from 62.1 to 79.3 percent. The adjusted ratios also rose considerably over this period, from 71.1 to 82.1 percent in the human capital specification and from 79.4 to 91.6 percent in the full specification. However virtually all of these gains occurred in the 1980s. This means that, while a reduction in the residual or unexplained gap played an important role in the narrowing of the gender wage gap over the 1980s, it has not been a factor since then (see also Blau and Kahn 2006). Figure 2 also indicates that the difference between the human-capital adjusted ratio and the unadjusted ratio fell dramatically over the 1980-2010 period, reflecting women’s increasing human capital levels relative to men’s. By 2010, the human capital variables (and the other variables included in this specification) explained very little of the gender wage gap: the unadjusted ratio was 79% compared to the adjusted ratio of 82%. As Goldin (2014) has commented, “As women have increased their productivity enhancing characteristics and as they ‘look’ more like men, the human capital part of the wage difference has been squeezed out.” As we shall see shortly in Table 4, this represents to some extent countervailing factors: women are now better educated than men but they continue to lag (slightly) in actual labor market experience. In the full specification, the adjusted ratio (91.6 percent) remained considerably higher than in the human capital specification (82.1 percent) in 2011, suggesting a continued substantial role for occupation and industry in explaining the gender wage gap (recall that union differences have now been virtually eliminated). Table 4 provides further detail on the contribution of particular labor market characteristics to the gender wage gap. Specifically, it shows the fraction of the total gender wage gap in 1980 and 2010 accounted for by gender differences in each group of variables for both the human capital and full specifications, again based on the Oaxaca-Blinder decomposition. The entries are the male-female differences in the means of each variable multiplied by the corresponding male coefficients from the current year wage regression. In Panel A, one sees the contribution of traditional human capital variables—education and experience—not controlling for industry, occupation or union status. This specification in effect allows human capital to affect these intervening variables and thus gives the reduced form effect of education and experience in explaining the gender wage gap. In 1980, the male advantage in education raised the gender wage gap somewhat, while the male experience gap contributed substantially (0.114 log points) and accounted for nearly a quarter of the gap. By 2010, due to the education reversal, women’s higher level of education slightly raised their relative wage. Moreover, the much smaller (compared to 1980) male advantage in labor market experience contributed only a small amount 0.037 log points to the gender wage gap, accounting for 16% of the now much reduced gender wage gap. Together, human capital factors (education and experience) accounted for 27% of the gender wage gap in 1980 compared to only 8% in 2010. Another notable change was the decline in the unexplained gap—from 0.341 log points in 1980 to 0.197 log points in 2010. This also contributed substantially to the narrowing of the gender gap over the period, although, as we have seen, the decrease in the unexplained gap occurred only during the 1980s. Nonetheless, unexplained factors accounted for a substantial share of the gender gap in both years, actually a bit larger share of gap in 2010 (85%) than in 1980 (71%). Table 4, Panel B, shows the decomposition of the gender pay gap using the full specification. Interestingly, the effects of education and experience are quite similar to that in Panel A, implying that the impact of these measures of human capital operates primarily within industries, occupations and union coverage status. In 1980, gender gaps in industry and occupation together accounted for 0.097 log points, or 20% of the gender pay gap, with gender differences in union coverage contributing an additional .03 log points or 6 percent of the gap. By 2010, the convergence in male and female unionization rates had virtually eliminated the contribution of this factor, but occupation and industry continued to account for a substantial gender gap of .117 log points or 51% of the smaller 2011 gender gap. Indeed, whether taken analyses using 1980 age weights, we found that the overall female to male wage ratio would have been 80.7% in 2010, compared to its actual value of 79.3% as shown in Table 1 and Figure 2, a slight increase as expected. However, the adjusted ratios were very similar to those shown in Figure 2.
7 separately or combined, occupation and industry now constitute the largest measured factors accounting for the gender pay gap. In both years, the unexplained gap was considerably smaller in the full specification than in the human capital specification, also highlighting the importance of industry and occupation. As in the case of the human capital specification, a marked decline in the unexplained gap (from 0.231 log points in 1980 to 0.088 log points in 2010) contributed to the narrowing of the gender wage gap, and, again, this decrease occurred over the 1980s. However, as in the case of the human capital specification, unexplained factors continue account for a substantial share of the gender gap in 2010 (38%) as they had in 1980 (49%). The continued importance of occupation and industry in accounting for the gender gap, and the rise in the relative importance of these factors, suggests that future research on explanations might fruitfully focus on gender differences in employment distributions and their causes. This meshes well with increased attention to the role of firms as firm-worker matched data increasingly become available. One puzzling finding in Table 4 is that, despite the occupational improvements of women shown in Table 3, gender differences in occupation accounted for a larger pay gap in 2010 than in 1980 (0.076 vs. 0.051 log points). However, while women upgraded their occupations during this period, the wage consequences of gender differences in occupations became larger as well. We study these consequences formally in Table 5. There we provide estimates of the impact of changes in the gender gaps in covariates on the change in the gender wage gap using a constant set of male wage coefficients (for 1980 or 2010). To do this we adapt an approach developed by Juhn, Muphy and Pierce (1991) (see also Blau and Kahn 1997), which also yields estimates of the effect of changing coefficients and the effect of changes in the unexplained gap. We begin with male (m) wage and female (f) wage equations as in (1) and (2) above for each of the two years (0, 1). Then,
(4) Effect of Changing Means = ∆ ∆
(5) Effect of Changing Coefficients = ∆
(6) Effect of Changing Unexplained Gaps = where and have been defined previously and a Δ prefix signifies the (mean) male-female difference for the variable immediately following. The effect of changing means measures the contribution of changes in male-female differences in measured labor market characteristics (X’s) on changes in the gender wage gap. So, for example, if women move into higher paying occupations it will reduce the gender wage gap. The effect of changing coefficients reflects the impact of changes in prices of measured labor market characteristics, as indexed by male coefficients, on changes in the gender wage gap. For example, given that women are located in different occupations than men, an increase in the return to occupations in which men are more heavily represented weights the gender difference in occupations more heavily and hence raises the gender wage gap, all else equal. Finally, the effect of changing unexplained gaps measures the impact of this factor on changes in the gender wage gap, with, e.g., a declining unexplained gap working to decrease the gender wage gap. The impact of changing means, changing coefficients, and changes in the unexplained gap together sum to the observed change in the total wage gap. The first two columns of Table 5 use the 1980 Male Wage Equation and 2010 Male-Female differences in the means of the covariates as the base, while the second two columns use the opposite values as base, in each case chosen to exhaust the explained portion of the change in the gender pay gap. In the human capital specification (giving the largest estimate of the impact of these variables) , women’s improvements in education and experience taken together are shown to narrow the gender pay gap by 0.092 to 0.098 log points, or about 38-40% of the actual closing of the gender pay gap. Thus,
8 improvements in these traditional measures of human capital were a very important part of the story explaining the decrease in the gender pay gap. Results for the Full Specification illuminate the role of industry, occupation, and unionism. Taken together these variables narrowed the gender gap by .064-.066 log points or 26%-27% of the closing. This reflects convergence in men’s and women’s occupations and union status in roughly equal measure, with relatively little evidence of narrowing of industry differentials. In terms of occupational convergence, women reduced their concentration in administrative support and service jobs, relative to men, and, as we have seen, increased their representation in managerial and professional jobs, including traditionally male professions. As well as occupational upgrading of women, the female relative gains reflect some adverse trends for men, including the decline in their employment in production jobs and the increase in their employment in service positions, as well as their considerably larger loss of union employment. In both specifications, the decline in the unexplained gender wage gap plays a substantial role in accounting for the wage convergence of women and men, explaining 58% of the closing.15 (As we have noted previously, this decrease occurred almost entirely in the 1980s.) Of course this begs the question as to what caused this decrease. There are a number of possible sources. The two most straightforward are that the decline represents a decrease in discrimination against women and/or a decrease in gender difference in unmeasured characteristics. Also potentially important are demand shifts favoring women relative to men and trends in the extent and type of selection of women and men into the labor force. In Blau and Kahn (2006), we present some evidence consistent with each of these possible explanations, suggesting that all might have played role. These are all issues that we address below. The decomposition presented in Table 5 also permits us to identify the role of changes in overall prices (coefficients) in affecting the trends. In general, for the 1980-2010 period, price changes are not found to play a major role in the human capital specification, but adverse price movements did negatively affect women’s gains in the full specification, almost entirely due to rising returns to occupations in which women were underrepresented. However, female improvements in the explanatory variables and a narrowing of the unexplained gap more than outweighed these adverse price changes. This analysis highlights the notion that shifts in labor market prices can affect women’s progress in narrowing the gender wage gap. The role of wage structure in affecting changes over time in relative wages of women, as well as differences across countries in the magnitude of the gender wage gap, is considered in Section 6. 2.2 Explaining the Gender Wage Gap Across the Wage Distribution As we saw in Table 1, as of 2010, (i) there was a relatively large gender gap at the top of the distribution and (ii) the wage gap fell more slowly over the 1980-2010 period at the top than at other portions of the distribution. These two patterns suggest the notion of a “glass ceiling” in which women face barriers in entering the top levels of the labor market and which we discuss in more detail in Section 3. To provide some further evidence on this phenomenon, we decompose the gender pay gap at specific percentiles of the distribution into portions due to covariates and portions due to wage coefficients. The latter component corresponds to the unexplained gap and, while as noted above, is sometimes taken to be a measure of discrimination, may be a biased estimate. To study the unexplained gap across the distribution, we use a method developed by Chernozhukov, Fernández-Val and Melly (2013) which decomposes unconditional intergroup gaps (in our case, male-female gaps) at a given percentile into a portion due to the distribution of characteristics and a portion due to different wage functions conditional on characteristics. This latter portion corresponds to the unexplained gap. As discussed by the authors, the method involves computing the
15 Coincidentally, the female residual fell by almost identical amounts in the Human Capital and Full Specifications (0.1432-0.1433 log points). Using the female coefficients as the base yielded qualitatively similar results for the changes in characteristics; however, the effects of prices changes were very small.
9 distribution of characteristics and the conditional wage distribution by gender. For example, as above, let log wages be denoted by Y, y be a specific value of log wages, m represent males, f represent females, and X be a vector of characteristics affecting wages. Then,